I am trying to use chain rule in derviating the loss for the Softmax function, but i stuck. In this great answer: I can't realize, why the log(p_k) derivative with respect to o is 1/p_k but not 1/(p_k * ln(10))?
Derivative of Softmax loss function issue
0
$\begingroup$
linear-algebra
derivatives
machine-learning
-
0In the post you've referenced, the natural logarithm is being used, so you don't need the conversion factor for a base-10 log. Keep in mind that in many (most?) branches of applied math and science, the natural logarithm is denote as $\log(x)$. It's mainly in undergraduate courses where the ${\rm ln}(x)$ notaton is encountered. – 2017-02-25
-
0@greg, thanks a lot! – 2017-02-26